Reproduce “OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yu Zhou, Cheng-Fu Yang, Rui Sun, Di Wu, Sibo Peng, Mingyang Zhang, Zhen Yang, Ruiyi Zhang and 3 more
OpenVLHarness equips multimodal language models with visual and information tools, rendered tool outputs, and reusable multimodal memory, allowing them to gather and combine evidence rather than rely only on the input image. Across 22 benchmark datasets, Qwen3-VL-8B’s unweighted mean score rises from 34.24 without tools to 45.41 with the system. This is not pooled accuracy, and crowded-scene counting remains difficult.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.